Syafiie Syamaun

Professor in Process Control

Process Control Class

This page is under construction

Process Dynamic and Control covers

  1. Introduction to Process Dynamics and Control
  2. First Principle Mathematical Modeling
  3. Dynamical Systems: Dynamical behaviour
  4. Feedback
  5. Basic PID Controller Design
  6. Controller Structure
  7. Frequency Domain Analysis
  8. Systems with Multiple Input and Multiple Output
  9. PID controller Design for MIMO
  10. Discrete-time and sequence contro
  11. Machine Learning for Control

Introduction to Process Dynamics and Control

First Principle Mathematical Modeling

The objectives are to:

Before we begin discussion on dynamic modeling, we first state the deffinition of "mathematical model" and "dynamic". In the year 1974, Eykhoff had defined that a mathematical model is a representation of the essential aspect of existing system (or a system to be constructed) which represent knowledge of that system in a usable form.

In its simplest form, a mathematical model is nothing more than equations that relate the values of variables, it can be decomposed into four basic components: input variables, output variables, parameters, and operators. Output variables are the unknown quantities that the model is designed to deduce from values of the known inputs. The operators in the model define the mathematical manipulations required to compute the value of the output from the inputs and parameters.

The dynamic is ( Of or relating to energy or to objects in motion, Merriem Webster Dictionary) the time varying behavior of processes. Most of chemical process behaviors are dynamically continuous changes respecting to inputs. The process in steady state change indicates where the process is going and the dynamic characteristic of the process indicates what dynamic path it will take. The dynamic of the process is usually represented by difference, differential, parametric, or state space models. In the following, these dynamical models are discussed.

A dynamic mathematical model can be represented in difference and differential equations. Usually, the model developed based on first principle is differential equations, they can be lumped or distributed equations. A model developed from a given input-output data is difference equations, which are usually represented in parametric model.

Difference equation

Suppose there is defined a sequence of points, perhaps representing discrete equally space time point, indexed by \(k\). Suppose also that there is a value \(y(k)\) (a real number) associated with each of these point. A difference equation is an equation relating the value \(y(k)\), at point \(k\), to value at other (usually neighboring) points. A simple example is \[y(k+1)=ay(k) \qquad k=0,1,2,\dots \] Difference equation may, however, be much more complicated than this. A solution of a difference equation is a function \(y(k)\) that reduces the equation to an identity. For example, corresponding to the first-order equation above, the function \(y(k)=a^k\) reduces the equation to an identity, since \(y(k+1)=a^{k+1}=aa^k=ay(k)\).

Usually, the first order difference equation is represented to take into account the bias of the function, which is known as the error compensate of the function. The equation is \[y(k+1)=ay(k)+b\] linear, has a constant \(a\), and a constant forcing term \(b\). The solution of the simple representation of the identity equation is by finding the simplest relation and correlation of the value in the neighborhood. From the equation above, the value of \(y\) at an initial point \(k_0\), say \(k_0=0\) and specify \(y(0)=y_0\). The following corresponding values are depend on the initial value, which in sequence is represented as \[ \begin{align} y(k_0=0)=&y_0\\ y(k_1=k_0+1=1)=&ay(0)+b=ay_0+b\\ y(k_2=k_1+1=1+1=2)=&a(y_1)+b=a(ay_0+b)+b=a^2y_0+ab+b\\ \vdots\\ y(k_n)=&a^ny_0+(a^{n-1}+a^{n-2}+\dots+a+1)b \end{align} \] The first term of the right hand side of the equation depends on the initial value, \(y_0\). However, the second term corrects the sequence of errors. For \(a=1\), the expression reduces simply to \[y(k_n)=y_0+nb.\] However, in the real facing problem, the weighted parameter is not necessarily equal to 1. Therefore, if \(a\neq1\), the expression can be somewhat simplified by collapsing the geometric series using \[ 1+a+a^2+\dots+a^{n-1}=\frac{1-a^n}{1-a} \] Therefore, the desired solution in closed-form is \[ y(k_n)=\left\{ \begin{array}{ll} y_0+nb, & a=1 \\ a^ny_0+\frac{1-a^n}{1-a}b, & a\neq1 \end{array} \right. \]

Differential equation

Suppose there is an interval, \(t_0\leq t\leq t_1\), representing an interval of continuous time. Suppose also that there is a value \(y(t)\) associated with each point \(t\) in that interval. Then \(y(t)\) is a function defined on the interval. A differential equation is an equation connecting such a function and some of its derivatives. A simple example is the equation \[\frac{dy}{dt}=ay,\quad y(0)=y_0\] which is an autonomous system. To drive the system is by introducing a force variable \(b\) such as \[\frac{dy}{dt}=ay+b,\quad y(0)=y_0\] The solution for the autonomos system is \[y(t)=y_0e^{at} \] and solution for non-autonomous system is given \[ y(t)=-\frac{b}{a}+\left(y_0+\frac{b}{a}\right)e^{at} \]

Phenomena Law

In general the system can be modeled by arranging that the system is influenced by a receiving and releasing material or energy and/or system's characteristics. This can be a chemical reaction or something else. \[\begin{bmatrix} \text{Rate of} \\ \text{Accumulation} \end{bmatrix}=\begin{bmatrix} \text{Rate of Entering} \\ \text{the systems} \end{bmatrix}-\begin{bmatrix} \text{Rate of Leaving} \\ \text{the systems} \end{bmatrix}+\begin{bmatrix} \text{Rate of Generation } \\ \text{or Consumption within system} \end{bmatrix}\]

Laplace Transformation

Dynamical Systems: Dynamical behaviour

Feedback

Here come text

Basic PID Controller Design

\[ u(t)=k_c\left(e(t)+\frac{1}{t_i}\int_0^te(\tau)d\tau+t_d\frac{de(t)}{dt}\right) \] where

Thus, it can be presented in Laplace transform as \[\begin{align} U(s)=&k_c\left(E(s)+\frac{E(s)}{t_is}+t_ds E(s)\right)\\ =&k_c\left(1+\frac{1}{t_i s}+t_ds\right)E(s) \end{align} \]

Ziegler-Nichols Tuning Method

Take a first-order plus dead time (FOPDT) plant \[G_p(s)=\frac{k_p}{\tau_ps+1}e^{-ls}\] apply a unit step \(\Delta u=1\) and \(t=0\). The time-domain output (for \(t\ge l\)) is \[y(t)=k_p\left(1-e^{-\frac{t-l}{\tau_p}}\right),\quad t\ge l\] and \(y(t)=0 \) for \( t < l \). Let for example the plant has \(k_p=2, \tau_p=5 \)s and \( l=1 \)s, therefor \(t\ge 1\) is= \[y(t)=2\left(1-e^{-\frac{t-1}{5}}\right) \] The classical reaction-curve (open-loop) method uses the tangent at the beginning of the response. For the FOPDT model above:

  1. Slope at \(t=l^+\) (immediately after delay) is \[S=\left.\frac{dy}{dt}\right|_{t=l}=\frac{k_p}{\tau_p} \]
  2. The tangent line through \(t=l, y=0\) with slop \(S\) is \[y_{tan}(t)=S(t-l)\]
  3. That tangent reaches the final value \(k_p\) at time \(t=l+\tau_p\) because \(S\tau_p=\frac{k_p}{\tau_p}\tau_p=k_p\). Thus the tangent intercepts give \(l\) and \(l+\tau_p\) directly, so visually/readable: \[\mbox{read off}\; l, \tau_p=t_{tangen\; at\; final}-l \]
For the example, it gives \(S=0.4\), tangen line \(y_{tan}​(t)=0.4(t-1) \). Set \(y_{tan}=k_p=2 \) to find he intercept: \(0.4(t-1)= 2\Rightarrow t-1=5\Rightarrow t=6 \) which equals \(l+\tau_p=1+5 \) So from the observed step response you would extract: \[k_p=2,\tau_p=5 s,l=1 s \] The classic Ziegler–Nichols open-loop (reaction-curve) PID tuning rules give controller parameters in terms of \(k_p,\tau_p\) and \(l\) One convenient form uses \[R\equiv\frac{k_p}{\tau_p} \] (the initial slope). Common formula set (open-loop Z-N) for PID is: \[k_c=\frac{1.2}{Rl},\quad t_i=2l,\quad t_d=0.5l \] for the example, it gives \[k_c=\frac{1.2}{0.4\times 1}=3,\quad t_i=2\times 1=2,\quad t_d=0.5\times 1=0.5 \] Thus it gives \[U(s)=3\left(1+\frac{1}{2s}+0.5s\right)E(s)\]

frequency

Controller Structure